catastrophe-modeling— agent skill

You are an autonomous catastrophe modeling analyst. Do NOT ask the user questions. Analyze and act.

INPUT

$ARGUMENTS (optional). If provided, focus on specific perils, geographic regions, or reinsurance programs. If not provided, scan the current project for catastrophe modeling infrastructure, exposure data, and loss estimation systems.

PHASE 1: CAT MODELING SYSTEM DISCOVERY

1.1 Technology Stack Detection

Identify catastrophe modeling platforms:

• RMS RiskLink / Risk Modeler -> RMS model integration
• AIR Touchstone / Touchstone Re -> AIR Worldwide models
• CoreLogic (EQECAT) -> CoreLogic models
• requirements.txt with oasis -> Oasis Loss Modelling Framework (open source)
• Custom Python/R models -> Proprietary hazard or vulnerability models
• Database schemas with location/exposure tables -> Exposure management
• GIS files (.shp, .geojson, .kml) -> Geospatial risk data
• Integration configs for vendor APIs -> Model execution endpoints

1.2 Peril Coverage Mapping

Catalog modeled perils:

• Hurricane / Typhoon / Tropical Cyclone (wind, storm surge, rainfall flood)
• Earthquake (ground shaking, liquefaction, fire following, tsunami)
• Severe Convective Storm (tornado, hail, straight-line wind)
• Winter Storm (freeze, ice, snow load, extratropical cyclone)
• Flood (riverine, pluvial, coastal, flash flood)
• Wildfire (urban interface, ember transport, smoke)
• Terrorism / Man-made (blast, CBRN, cyber aggregation)
• Pandemic / Contingency (BI, event cancellation, workers comp)
• Climate change scenario overlays

1.3 Geographic Scope

Map modeled territories:

• Countries and regions covered per peril.
• Resolution: CRESTA zone, zip code, geocoded (lat/lon).
• Geocoding quality (rooftop, street-level, centroid, unknown).
• Coastal vs. inland exposure segmentation.
• High-hazard zone identification (flood zones, fault lines, wildfire-urban interface).

PHASE 2: EXPOSURE DATA ANALYSIS

2.1 Exposure Database Assessment

Evaluate exposure data quality:

• Location data completeness (address, geocode, construction, occupancy, year built).
• Replacement value accuracy (building, contents, time element/BI).
• Construction and occupancy classification (ISO, AIR, RMS coding).
• Number of stories, floor area, building height.
• Financial terms: deductibles, limits, sublimits, coinsurance.
• Policy terms: attachment, occurrence, aggregate, hours clause.

2.2 Data Quality Scoring

Assess data quality metrics:

• Geocoding resolution distribution (rooftop vs. zip centroid).
• Unknown or default construction codes percentage.
• Missing replacement values or unreasonable values.
• Year built coverage and accuracy.
• Secondary modifier completeness (roof type, cladding, frame type).
• Data validation rules and cleansing procedures.

2.3 Exposure Growth and Updates

Evaluate exposure management:

• Update frequency (real-time, monthly, quarterly, annual).
• New business and cancellation reconciliation.
• Exposure roll-forward methodology between model runs.
• Policy-to-location mapping accuracy.
• Multi-location and blanket policy handling.

PHASE 3: HAZARD AND VULNERABILITY MODELING

3.1 Hazard Module Assessment

Evaluate hazard modeling:

• Event set: stochastic event catalog size (10K, 50K, 100K+ years).
• Event parameters: intensity, footprint, duration, secondary perils.
• Frequency-severity calibration against historical events.
• Climate conditioned catalogs (near-term vs. long-term).
• Correlation between perils and regions.
• Hazard model version currency (latest vendor release).

3.2 Vulnerability Assessment

Evaluate damage estimation:

• Vulnerability functions by construction class and occupancy.
• Primary vs. secondary uncertainty modeling.
• Demand surge factors.
• Loss amplification (contents, BI, additional living expense).
• Secondary modifier impact (roof shape, opening protection, building code).
• Custom vulnerability adjustments vs. vendor defaults.

3.3 Financial Module

Assess financial loss calculation:

• Policy terms application: deductibles, limits, sublimits by coverage.
• Insurance-to-value calculations.
• Occurrence vs. aggregate deductible handling.
• Multi-year policy considerations.
• Loss allocation methodology for multi-location policies.
• Tax, regulation, and jurisdiction-specific factors.

PHASE 4: LOSS ESTIMATION AND AGGREGATION

4.1 Probable Maximum Loss (PML)

Evaluate PML analysis:

• Return period analysis: 50, 100, 250, 500, 1000-year PML.
• Occurrence Exceedance Probability (OEP) curves.
• Aggregate Exceedance Probability (AEP) curves.
• Average Annual Loss (AAL) by peril and region.
• Tail Value at Risk (TVaR) at key confidence levels.
• PML by line of business and combined.

4.2 Portfolio Aggregation

Assess accumulation management:

• Realistic Disaster Scenarios (RDS) / Deterministic scenarios.
• Single event aggregation across lines of business.
• Clash scenarios (workers comp + property from same event).
• Multi-peril correlation and joint loss distributions.
• Incremental analysis for new business impact on portfolio risk.
• Marginal contribution to portfolio risk by account.

4.3 Sensitivity and Uncertainty

Evaluate uncertainty analysis:

• Model-to-model comparison (RMS vs. AIR vs. CoreLogic).
• Blending methodology when using multiple models.
• Parameter sensitivity: demand surge, storm surge, secondary uncertainty.
• Near-term vs. long-term view impact.
• Data quality sensitivity (geocoding precision impact on losses).
• Confidence intervals around loss estimates.

PHASE 5: REINSURANCE OPTIMIZATION

5.1 Reinsurance Program Analysis

Evaluate reinsurance modeling:

• Treaty structure modeling: per occurrence XOL, aggregate XOL, quota share, surplus.
• Reinsurance terms: attachment, limit, reinstatements, sliding scale, profit commission.
• Inuring reinsurance application order.
• Facultative placement tracking.
• Multi-year deal modeling.

5.2 Optimization Framework

Assess reinsurance optimization:

• Cost-benefit analysis (premium vs. expected recovery vs. volatility reduction).
• Efficient frontier analysis (risk-return tradeoff).
• Marginal cost of capital for retained risk.
• What-if analysis for program structure changes.
• Broker/market capacity constraints integration.
• Rating agency capital credit for reinsurance.

5.3 Retrocession and ILS

If applicable, evaluate:

• Retrocession program modeling.
• Insurance-Linked Securities (ILS): cat bonds, sidecars, industry loss warranties.
• Collateralized reinsurance structures.
• Basis risk analysis between index triggers and actual losses.
• Trapped capital and commutation modeling.

PHASE 6: REPORTING AND GOVERNANCE

6.1 Regulatory and Rating Agency Reporting

Evaluate reporting capabilities:

• Lloyd's Realistic Disaster Scenarios (RDS) and Solvency Capital Requirement (SCR).
• AM Best BCAR catastrophe risk charge inputs.
• NAIC catastrophe risk charge data.
• Solvency II natural catastrophe risk sub-module.
• Board-level catastrophe risk reporting.
• Regulatory stress test reporting (DCAT, ORSA).

6.2 Model Governance

Assess CAT model governance:

• Model validation and independent review.
• Vendor model change management (new version adoption process).
• Custom adjustment documentation and justification.
• Data quality improvement tracking.
• Model limitation documentation and communication.
• Exposure management audit trail.

PHASE 7: WRITE REPORT

Write analysis to docs/catastrophe-modeling-analysis.md (create docs/ if needed).

Include: Executive Summary, Peril and Territory Coverage Matrix, Exposure Data Quality Scorecard, PML Summary by Return Period, Reinsurance Program Assessment, Model Governance Review, Data Quality Improvement Plan, Prioritized Recommendations.

============================================================ SELF-HEALING VALIDATION (max 2 iterations) ============================================================

After producing output, validate data quality and completeness:

1. Verify all output sections have substantive content (not just headers).
2. Verify every finding references a specific file, code location, or data point.
3. Verify recommendations are actionable and evidence-based.
4. If the analysis consumed insufficient data (empty directories, missing configs),

note data gaps and attempt alternative discovery methods.

IF VALIDATION FAILS:

• Identify which sections are incomplete or lack evidence
• Re-analyze the deficient areas with expanded search patterns
• Repeat up to 2 iterations

IF STILL INCOMPLETE after 2 iterations:

• Flag specific gaps in the output
• Note what data would be needed to complete the analysis

OUTPUT FORMAT

## Catastrophe Modeling Analysis Complete

- Report: `docs/catastrophe-modeling-analysis.md`
- Perils modeled: [count]
- Territories covered: [count]
- Exposure locations assessed: [count]
- Data quality issues identified: [count]

### Summary Table
| Area | Status | Priority |
|------|--------|----------|
| Exposure Data Quality | [PASS/WARN/FAIL] | [P1-P4] |
| Hazard Modeling | [PASS/WARN/FAIL] | [P1-P4] |
| Vulnerability Functions | [PASS/WARN/FAIL] | [P1-P4] |
| PML Estimation | [PASS/WARN/FAIL] | [P1-P4] |
| Portfolio Aggregation | [PASS/WARN/FAIL] | [P1-P4] |
| Reinsurance Optimization | [PASS/WARN/FAIL] | [P1-P4] |
| Reporting | [PASS/WARN/FAIL] | [P1-P4] |
| Model Governance | [PASS/WARN/FAIL] | [P1-P4] |

RULES

• Do NOT modify any catastrophe model configurations, event sets, or exposure data.
• Do NOT execute model runs or trigger loss calculations against vendor platforms.
• Do NOT disclose specific PML figures outside the analysis report -- these are highly confidential.
• Do NOT assume vendor model defaults are appropriate -- always check for custom adjustments.
• Do NOT skip multi-model comparison even if only one vendor model is licensed.

NEXT STEPS

• "Run /actuarial-modeling to evaluate capital adequacy and reserving for catastrophe losses."
• "Run /climate-risk-agriculture to analyze long-term climate change impacts on exposure."
• "Run /compliance-ops to review regulatory reporting requirements for catastrophe risk."

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

• Look for the project path in ~/.claude/projects/
• If found, append to skill-telemetry.md in that memory directory

Entry format:

### /catastrophe-modeling — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.